thoughtworks/arithmetic-sorl-data
Arithmetic SoRL Data Training and evaluation data for the SoRL Arithmetic Interpretability Study. Small transformers trained on integer addition/subtraction, with SoRL to externalize carry/borrow circuits as explicit abstraction tokens. Reference: Quirke et al., "Understanding Addition and Subtraction in Transformers" (2024). Paper: arXiv:2402.02619 — see Table 8 for complexity classification and Section 3 for sub-task definitions. Dataset Structure… See the full description on the dataset page: https://huggingface.co/datasets/thoughtworks/arithmetic-sorl-data.
0125
1{2 "n_digits": 6,3 "ops": "add",4 "train_size": 500000,5 "val_size": 10000,6 "enrichment": "sum-to-9 (addition) + equal-digit borrow (subtraction) + 1% equal operands",7 "columns": [8 "tokens",9 "labels",10 "op",11 "complexity",12 "cascade_depth",13 "x_digits",14 "y_digits",15 "z_digits",16 "sa",17 "st",18 "sv"19 ]20}